Load testing baseline with k6

[Yentec]

Context

The architecture (Redis cache + async tracking) is designed for throughput on the redirect hot path, but there's no published benchmark to back the claim.

Goal

Establish a documented latency/throughput baseline for GET /:code and publish results in the README.

Approach

• Add tests/load/redirect.js: a k6 script with stages (ramp 0 -> 100 -> 0 over 1 min).
• Run against the production instance once warmed up.
• Capture p50, p95, p99 latency and req/s.
• Add a small section "Performance" to the README with the numbers and the test command.
• Write a short ADR explaining the result vs the architecture choices.

Acceptance criteria

• tests/load/redirect.js committed
• README "Performance" section with reproducible command
• ADR 0006 — "Redirect latency baseline"

Out of scope

Continuous load testing in CI. Render free tier sleeps, so CI runs would be noisy. Manual runs only.